Fusion mattes adding additive line.

Hey!

Working on a show that's using a lot of split screens and wipes. I'm using Fusion to create these mattes, using the matte control node and piping a rectangle shape into the garbage matte, which I can then animate as needed.

However, it's adding a one pixel bright glow to the borders. Mattes sizes are set to .33 on the left and .34 in the middle, plus .33 on the right. So they're not physically overlapping each other.

Trying to figure out what's causing this behavior, and how to stop it from doing it. I'm sure there's plenty of work arounds, but wanting to understand why this workflow is creating this issue.

Using Resolve Studio 18.5 on Mac OS 12.7.4 - 2.5 GHz 28-Core Intel Xeon W with 512 GB 2933 MHz DDR4 memory and a AMD Radeon Pro Vega II Duo 32 GB graphics card.

Here's a great example of what I'm talking about.

The first image (black background) has a black background merged to the shots in fusion, and the lines have disappeared.

The second image (alpha background) is just the same shot with an alpha background. Once back in the edit pages, there's obvious lines all around the edges.

You'll probably need to open the images in separate tabs and compare to see the difference.

This isn't a viable work around for a lot of the shots I'm doing, as they're multiple layers going on, which I'm keeping separate for color.

These lines aren't visible in Fusion's window, so what's causing them to appear back in the edit page? Is there a separate alpha setting that I should be aware of in the clip inspector?

Change the rescaling filter in Resolve project settings to bilinear. It looks like another case of sharpening rescale filter messing with semitransparent areas.

Messed with the different sharpening settings, and it slightly changes it, but doesn't remove it.

Any other thoughts?

Can you share your comp for inspection?

If you probe the pixel values on that line, what are the RGBA channel values on either element before the merging and after merging in Fusion page? Does it have solid 1.0 alpha after merging?

Trying to figure out what's causing this behavior, and how to stop it from doing it. I'm sure there's plenty of work arounds, but wanting to understand why this workflow is creating this issue..

I tried to do what you did with mat control nodes and all that, but I was not able to replicate your problem. But the first thing I've noticed is nodes set up that seemed overly complicated for what you said you wanted to do. Of course I don't know if there is something you haven't said that you want and need all those nodes. But I think I would appraoch it in a simpler way and reduce changes of something going wrong. Whatever it is.

I would simply use the rectangle mask as you do and plug it into transform node and set transform node to multiply by mask. Those two nodes should do the same as you have been trying with four nodes. Should speed processing and set up. Unless you want something I'm not aware of it should simplify the process.

sshot-4087.jpg (9.77 KiB) Viewed 815 times

sshot-4086.jpg (212.18 KiB) Viewed 815 times

Regarding your problem in the edges you mentioned that you only see in edit page, I am unable to replicated it. But perhaps there is some setting that is different between us that could be the culprit.

I'm no Windows 10. DaVinci 18.6.6 Studio.

Got me thinking. Could it be related to GPU drivers or something? Do you use the latest stable ones?

Thanks for the thoughts!

Nice trick with the transform node, Kruno. Hadn't realized that function.

For a lot of things, it definitely would simplify the effect, but there's a lot of different effects going on. I'll keep that in the back of my mind.

I switched one clip over to using the transform node only, instead of piping through a matte control, then a transform node, and the issue is still visible. So definitely seems to be something relating to the environment I'm in.

As for getting RGBA values, what's the easiest way to do that on a Mac? I've never had to find that information before.

I'll try and get a project uploaded here shortly.

So in setting up a new project, it's an ACES issue.

This project is in ACES, but when importing the same file, with the same Fusion node tree, to a new timeline in DaVinci YRGB, the line disappears.

Bypassing color management also removes the issues, as I would expect, but that will only help with clips that are already Rec 709, not anything RAW.

Any thoughts on that?

In that case I'd point my finger on something being messed up in how Resolve does the colorspace transformations in relation to alpha values. Did you probe that line in fusion for what the RGBA values are there and how they compare to values adjascent to the line?

Changing between color science doesn't change any of the RGBA values in Fusion, although there is definitely a difference in the previews between ACES and DaVinci RBG.

Would be great to figure out why this is the case, as it makes for a cumbersome workflow to create these sorts of effects using Fusion while in ACES.

Changing between color science doesn't change any of the RGBA values in Fusion

Problem appears when going to the edit page, doesn't it? And there are transformations happening during that transition from Fu page to Edit/Color. If you have lets say RGBA value set there where alpha isn't full value, something in these lines will happen: RGB channels get divided by alpha value, then color transforms are applied, then result is multiplied with alpha again. Depending on what exactly is done, having alpha as not 1.0 can obviously have implications on how the result looks like. So again, what are those RGBA channel values in Fu page on some pixel on that line and adjascent pixel where no change is produced. Do you have alpha values that are lower than 1.0 on that line area?

There's definitely a difference between the preview in the Fusion page between ACES and DaVinci RGB. So I would believe that it's an issue with how ACES is managing the alpha channel within Resolve, not just on the edit page.

Sorry, the RGBA values were on those screenshots I uploaded, just small in the bottom left corner. The values are the same between ACES and DaVinci RGB though.

A = 0 where the image is transparent and A = 1 when it's over the image.

Between 0 and 1, I'm getting values of A = 0.001144 and A = 0.56699 as I move the cursor up from transparent to the image. Again, those values are the same regardless of whether I'm in ACES or DaVinci RGB.

If I apply a black background within Fusion, there are no lines visible. It's only displaying when there's transparency within Fusion.

Does that help?

There's definitely a difference between the preview in the Fusion page between ACES and DaVinci RGB. So I would believe that it's an issue with how ACES is managing the alpha channel within Resolve, not just on the edit page.

Sorry, the RGBA values were on those screenshots I uploaded, just small in the bottom left corner. The values are the same between ACES and DaVinci RGB though.

A = 0 where the image is transparent and A = 1 when it's over the image.

Between 0 and 1, I'm getting values of A = 0.001144 and A = 0.56699 as I move the cursor up from transparent to the image. Again, those values are the same regardless of whether I'm in ACES or DaVinci RGB.

If I apply a black background within Fusion, there are no lines visible. It's only displaying when there's transparency within Fusion.

Does that help?

Just out of my own curiosity. What happens if you apply alpha multiply node? Does the problem still happen?

Happy for you to be as curious as you want here!

Adding the alpha multiply node reduces it some, but doesn't remove it completely.

I'm finding it odd that some clips are fine, and other clips aren't as well. Wondering if it's a certain brightness level that's tripping it out with the alpha values.

Thankfully, I'm only delivering 1080 masters for this show, and the lines aren't visible on those masters. But we're doing 4K archival masters, so would like to get it fixed before we do those exports.

Happy for you to be as curious as you want here!

Hehe thanks.

Adding the alpha multiply node reduces it some, but doesn't remove it completely.

I though maybe the semitransparent edges were not properly premultiplied.

I'm finding it odd that some clips are fine, and other clips aren't as well. Wondering if it's a certain brightness level that's tripping it out with the alpha values.

I suppose its possible, but I can't say without actually working with the files myself. However you can try maybe Brightness contract node and clip the black and whites. Just to be sure. Sometimes its helps.

sshot-63.jpg (26.87 KiB) Viewed 519 times

Thankfully, I'm only delivering 1080 masters for this show, and the lines aren't visible on those masters. But we're doing 4K archival masters, so would like to get it fixed before we do those exports.

Another thing maybe, if you are resizing footage, perhaps some algorithm for resizing is causing the issues. That is last thing I can think off. Usually in fusion and edit page you can choose various algorithms for resizing and some can at times cause issues, but its probably nothing. Worth the shot, though. "Linear" is default in Fusion and safest one.

Sorry for my delayed response. Got hit with a lot of deliveries and finishing!

I tried adding the brightness/contrast node and clipping the black and whites, but as you suspected, that didn't do anything.

It's also not a transform node issue, as it's happening on clips where I'm only adding a mask.

From what I can gather, it's an ACES problem with alpha in Fusion only. Any cropping or resizing done within the edit page is completely fine.

Adding the alpha multiply node should be enough to get me there, but I'm curious as to whether it's an ACES issue, or the way that Fusion is handling ACES color.

Adding the alpha multiply node should be enough to get me there, but I'm curious as to whether it's an ACES issue, or the way that Fusion is handling ACES color.

It could be. I don't know for sure. Oh, well, at least you are still keeping busy. Good thing.